A Hybrid Model for Cloud Resource Utilization Forecasting Using Machine Learning and Evolutionary Optimization

Le résumé fourni par la source

To manage cloud resources efficiently, reduce costs, and improve service delivery in dynamic cloud environments, accurate forecasting is essential. The highly variable and nonlinear nature of cloud workloads often makes traditional prediction models ineffective. In this paper, Recurrent Neural Networks (RNNs) are integrated with evolutionary optimisation algorithms, specifically Genetic Algorithms (GAs) and Particle Swarm Optimisation (PSO), to develop a hybrid forecasting methodology. Combining GA-PSO enhances the prediction accuracy for both univariate and multivariate resource usage, with a focus on CPU and memory usage. According to experiments, FLPSONN outperforms FLGANN, FLNN, and FLPSONN in terms of mean absolute error (MAE) and percentage improvement when used with real-world cloud workload datasets. Hybrid models enable scalable and adaptive forecasting of cloud resource demands, thereby improving the efficiency and reliability of cloud service provisioning.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.